openFoam Hangout on Air #2 - Cloud Simulation, presentation by Dacolt
Download as PPTX, PDF2 likes1,211 views
Dacolt offers CFD modeling software and services for combustion applications. They have been involved in cloud computing for nearly a decade through their Tabkin combustion software and web services framework. An engineering project used OpenFOAM in the cloud to model a domestic burner and fatigue experiment, which was successfully completed on time using a cloud cluster. Dacolt's experience demonstrates that engineering simulations can feasibly use Open Source CFD in the cloud by selecting an appropriate cloud provider and being prepared to deal with related HPC issues.
openFoam Hangout on Air #2 - Cloud Simulation, presentation by Dacolt
- 1. Dacolt Combustion & CFD Open Source CFD in the cloud Google Hangout - Jan 2014 Document: Dacolt_Google_Hangout Revision: 1 © Dacolt 2009 - 2014 Dacolt is the trading name of Dacolt International BV, registered under number 14082157 at the Dutch chamber of commerce.
- 2. Contents • Introduction • Cloud activities @ Dacolt • Engineering project using Open Source CFD in the Cloud • Summary © Dacolt 2009 – 2014 www.dacolt.com 2
- 3. About Dacolt Dacolt offers software and services for CFD modeling of industrial combustion applications • Software & Services: – – – – CFD consultancy Software development Tabkin for CFD look-up tables (SaaS) Engineering services and training • Applications: – Internal combustion engines – Gas turbines – Industrial and domestic burners © Dacolt 2009 – 2014 www.dacolt.com 3
- 4. Cloud activities @ Dacolt • Tabkin: – Combustion software – Cloud based (SaaS) – Runs in Dacolt WSF Web Services Framework • Web Services Framework: – Create, manage & deploy HPC apps in the Cloud – Not only for combustion! © Dacolt 2009 – 2014 www.dacolt.com 4
- 5. © Dacolt 2009 – 2014 www.dacolt.com 5
- 6. Cloud activities @ Dacolt • Participation in UberCloud HPC Experiment since 2012 • Team expert for 5 teams: – – – – – Team Team Team Team Team 4: Supersonic flow (S&T), ESI, OpenFOAM®, Celia 36: Diesel engines (Dacolt), Ansys, Fluent, POD 52: GT burners (CSE Fire), ESI, OpenFOAM®, Bull XF 85: Power plants (OEM), ESI, OpenFOAM®, Bull XF 95: Combustion (Areté RC), ESI, OpenFOAM®, CPU 24/7 • Various experiences, hurdles and solutions – Mostly related to software installation or IT connection issues – Typically start-up problems © Dacolt 2009 – 2014 www.dacolt.com 6
- 7. © Dacolt 2009 – 2014 www.dacolt.com 7
- 8. Engineering project using Open Source CFD software in the Cloud • Project scope: – CFD analyses of various processes – Coupling to FE analyses for fatigue • Software: – ICON FOAMpro, now iconCFD – Derived from OpenFOAM – chtMultiRegionFoam derived solver • Computational resources: – Local cluster (32 cores – 2x Intel Xeon E5-2660 – Cloud cluster (96 cores – 6x Intel Xeon E5-2690) © Dacolt 2009 – 2014 www.dacolt.com 8
- 9. Domestic burner for hot water Burnt gas (approx. 1700 C) Hot water (80 C) Water Cold water (60 C) Cooled down gas (approx. 90 C) + condensate © Dacolt 2009 – 2014 www.dacolt.com 9
- 10. Fatigue experiment Burnt gas (approx. 1700 C) Hot water (80 C) Water Cold water (60 C) Cooled down gas (approx. 90 C) + condensate © Dacolt 2009 – 2014 www.dacolt.com 10
- 11. Fatigue experiment Burner stopped Flush ~ 60 s Water Cold water (15 C) © Dacolt 2009 – 2014 www.dacolt.com 11
- 12. CFD model setup • Gas-side CFD provides HT BC • Coupled water flow + conduction through tubes • Simulate fatigue experiment • 50M cells, 10 s, dt=4e-5 s • Runtime: – 6 weeks in office – Too long for customer deadline -> Scale up! © Dacolt 2009 – 2014 www.dacolt.com 12
- 13. Cloud resources • Target: – 2 weeks to find provider and start production runs – Reduce simulation turn-around time to 2 weeks • Main criteria: – Predictable scalability – Pricing • Sabalcore: – – – – Same CPU architecture (Intel Xeon Sandy Bridge) Same OS Same Infiniband interconnect Competitive pricing © Dacolt 2009 – 2014 www.dacolt.com 13
- 14. Cloud experience • Procedure: – – – – – – – Remote login via ssh from local workstation Upload mesh (50M cells) Remote partitioning Submit job to queue Remote recomposing Download intermediate results Local post-processing + transfer to FE analysts © Dacolt 2009 – 2014 www.dacolt.com 14
- 15. Cloud experience • Production run: – – – – 2 weeks on 96 cores ~ 32000 CPU-core hrs Nearly linear speed-up going from 32 to 96 cores 6,5 Gb of data per day, ~ 1 Tb in total Daily file transfer • Overall experience: – – – – Satisfying and applicable for engineering simulations Good to aim specific target platform for supplier selection Be aware of MPI-related issues and other HPC stuff Allow for sufficient time for file transfer (overnight) © Dacolt 2009 – 2014 www.dacolt.com 15
- 16. Summary • Dacolt involved in Cloud activities for nearly a decade: – Tabkin: cloud-based combustion software – WSF: PaaS for HPC applications – Team expert in various UberCloud HPC experiments • Engineering project using Open Source CFD in the cloud: – Surely feasible, especially if familiar with Linux – Set appropriate selection criteria for provider to minimise risks – Project achievements: • Delivered on time • Final fatigue results well in line with experiments © Dacolt 2009 – 2014 www.dacolt.com 16
- 17. Dacolt Combustion & CFD Open Source CFD in the cloud Google Hangout - Jan 2014 Document: Dacolt_Google_Hangout Revision: 1 © Dacolt 2009 - 2014 Dacolt is the trading name of Dacolt International BV, registered under number 14082157 at the Dutch chamber of commerce.